Redis: Chasing Timing Bugs Across the Stack

Today's merges were dominated by timing-related fixes — from flaky CI tests to a real data-loss window in cluster slot migration — plus a significant overhaul of the install and uninstall process.

Duration: PT2M45S

Episode overview

This episode is a short developer briefing from Redis.

It explains recent repository work in plain language.

  • Show: Redis
  • Published: 2026-08-27T13:15:53Z
  • Audio duration: PT2M45S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Good morning. It's August 27th, 2026, and this is your Redis briefing.

The clearest pattern in today's activity is timing. Multiple fixes landed to close gaps where code assumed something had finished when it hadn't — and in one case, that gap could actually lose writes.

Start with the serious one: PR 15629 from Roy Ben Moshe. During atomic slot migration, the source node could send a stream end-of-file signal for a handoff task without first checking whether the write-pause timeout had already expired. If it had, the source could resume accepting traffic before the destination…

That same theme — tests failing because of timing, not actual bugs — shows up twice more. PR 15715 fixed the reclaim-cache CI pipeline: a daemonized Redis server's parent process was exiting right after fork, before the child had finished TSC calibration and started listening, so test scripts connected too early and…

Separately, PR 15710 moved clock-calibration fallback logging out of raw standard error and into Redis's normal logging system. Since that clock code is linked into every Redis binary, unrouted stderr output risked being misread as a failure by test harnesses or tooling that treats any stderr…

On…

Nearby episodes from Redis

  1. Shoring Up CI and Config Safety
  2. Correctness Fixes and a Fast Clock Rollout
  3. Correctness Fixes and Command-Line Hardening
  4. Weekly Recap - A Coordinated Security Push Across Every Supported Line
  5. Correctness Bugs Sweep Across Core Commands
  6. TLS Hardening Takes Center Stage
  7. Streams Get Visibility, Filters Get Fixed
  8. Boundary Bugs and Cluster Ownership Cleanup